I am working on the project crack detection from the tongue and I am using yolov8n model . I have 900 images of tongues having crack on it . I have labeled this images using bouding boxex and trained the model . I am getting map50 0.58 and map50-95 0.25 . I want to increse the accuracy how can i increse it ? I have applied data augmentation technique and I incresed the data from 900 to 4500 images and trained the model then I got the map50-95 0.28 . I am new in this so i dont have any idea . Give me suggestion. Also I want the model should detect the crack if present and if no crack on the tongue it will no detect . So do i need to add the tongue images with no crack from improving the accuracy? and do i need to label the images with no crack ?
Do i need to add the images with no crack on the tongue and if i need to add this images then do i need to label them as well to get high accuracy . I want if in the input we send image with no crack then the model should not detect anything and if there is crack then it should detect crack .